Open-Omics-DeepVariant-V1.0

Open-Omics-DeepVariant-V1.0

This release of Open-Omics-DeepVariant is compatible with Google DeepVariant-V1.5.0 and up to 2x faster while achieving similar accuracy.

The main contributions of Open-Omics-DeepVariant include:

1. BF16 quantization for Call variant module
   - Applies BFLOAT16 quantization so as to use Intel® Advanced-Matrix-Extension (AMX), built-in hardware support for deep learning in latest Intel® fourth-gen scalable processors
   - Call variant module is ~2.7x faster than Open-Sourced version of DeepVariant
2. Parallel Post-process module:
   - Provides a parallel multi-processed implementation of post-process module
   - Using 56 CPU cores, up to 11x faster than DeepVariant’s sequential post-process module